codex-computer-use

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the official Playwright MCP server (@playwright/mcp) and the Playwright package from the npm registry to provide browser-based tools and visual capture capabilities.
  • [COMMAND_EXECUTION]: Executes host system utilities including screencapture, xcrun simctl, and npx playwright to interact with the desktop environment, mobile simulators, and web applications.
  • [PROMPT_INJECTION]: Inherits a risk of indirect prompt injection via the processing of external visual data. Adversarial instructions could be embedded within web pages or UI elements that the agent interprets while analyzing screenshots.
  • Ingestion points: Visual data captured from the screen, simulators, or browser sessions and passed to the agent via the -i flag.
  • Boundary markers: The skill does not provide specific delimiters or instructions to ignore commands embedded within the visual data.
  • Capability inventory: The skill utilizes workspace-write permissions, enabling file modifications and shell execution based on the results of visual analysis.
  • Sanitization: No automated sanitization or filtering of visual inputs is implemented before the content is passed to the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 03:17 PM
Security Audit — agent-trust-hub — codex-computer-use